The present invention comprises a decentralized coordination strategy called alpha-beta coordination. The alpha-beta coordination strategy is a family of collective search methods that allow teams of communicating agents to implicitly coordinate their search activities through a division of labor based on self-selected roles and self-determined status. An agent can play one of two complementary roles. An agent in the alpha role is motivated to improve its status by exploring new regions of the search space. An agent in the beta role is also motivated to improve its status, but is conservative and tends to remain aggregated with other agents until alpha agents have clearly identified and communicated better regions of the search space. An agent can select its role dynamically based on its current status value relative to the status values of neighboring team members. status can be determined by a function of the agent's sensor readings, and can generally be a measurement of source intensity at the agent's current location. An agent's decision cycle can comprise three sequential decision rules: (1) selection of a current role based on the evaluation of the current status data, (2) selection of a specific subset of the current data, and (3) determination of the next heading using the selected data. Variations of the decision rules produce different versions of alpha and beta behaviors that lead to different collective behavior properties.
|
10. A method of coordinating collective search by a plurality of agents, wherein the position of agent i at time k is denoted νi(k), comprising, on each agent:
a) Determining an individual status corresponding to the agent's probability of being near the search goal; b) Determining the status of and direction to neighboring agents; c) Determining the number of higher status agents, where a higher status agent is a neighboring agent with a higher status than the agent; d) Selecting a decision rule based on the number of higher status agents, then moving according to the selected decision rule.
1. A method of coordinating collective search by a plurality of agents, comprising, on each agent:
a) Determining an individual status corresponding to the agent's probability of being near the search goal; b) Determining the status of and direction to neighboring agents; c) Determining the number of higher status agents, where a higher status agent is a neighboring agent with a higher status than the agent; d) If the number of higher status agents is less than an alpha0 threshold, wherein the alpha0 threshold is greater than zero, then moving according to an alpha0 decision rule; e) If the number of higher status agents is less than an alpha1 threshold, wherein the alpha1 threshold is greater than zero, then moving according to an alpha1 decision rule; f) If the number of higher status agents is greater than or equal to the alpha1 threshold, then moving according to a beta decision rule.
3. The method of
a) Do not move; and b) Determining the direction to each neighboring agent with a lower status and moving in a direction derived from a combination thereof.
4. The method of
a) Do not move; and b) Determining the direction to a higher status agent and moving in a direction derived from the agent's current position and the higher status agent's current position.
5. The method of
6. The method of
7. The method of
11. The method of
12. The method of
13. The method of
14. The method of
15. The method of
a) determining whether the agent is one of the m highest status agents, where m is greater than one and less than or equal to the number of agents, and b) if so then selecting a higher status agent having position ν*(k), and moving according to νi(k+1)=νi(k)+μ[ν*(k)-νi(k)], where μ is a scaling factor, and c) if not then moving according to νi(k+1)=νi(k)+a(k)[vA(k)-νi(k)], where vA(k) is a vector of the positions of the higher status agents, a(k) is a vector with elements aj(k)=sj(k)/S(k), and
16. The method of
a) determining whether the agent is one of the m highest status agents, where m is greater than one and less than or equal to the number of agents, and b) if so then selecting one or more higher status agents and moving according to νi(k+1)=νi(k)+a(k)[vA(k)-νi(k)], where vA (k) is a vector of the positions of the higher status agents, a(k) is a vector with elements aj(k)=sj(k)/S(k),
and μ is a scaling factor, and
c) if not then moving according to νi(k+1)=νi(k)+a(k)[vA(k)-νi(k)], where vA(k) is a vector of the positions of the higher status agents, a(k) is a vector with elements aj(k)=sj(k)/S(k), and
17. The method of
18. The method of
a) Do not move; b) Moving in a direction derived from the agent's current position and a higher status agent's current position; and c) Moving in a direction derived from the agent's current position and a combination of the direction to and status of the higher status agents.
|
|||||||||||||||||||||||||
This invention was made with Government support under Contract DE-AC04-94AL85000 awarded by the U.S. Department of Energy. The Government has certain rights in the invention.
This invention relates to the field of coordinating multiple agents in collective searching, specifically coordinating mobile robots for collective searching.
Many challenging applications in robotics involve distributed searching and sensing by a team of robots. Examples include mapping minefields, extraterrestrial and undersea exploration, volcano exploration, location of chemical and biological weapons, and location of explosive devices. In general, such applications can involve rough terrain including obstacles, non-stationary and dilute search goals, deliberate interference with the searchers, and limited opportunities for human interaction with the searchers. Limited human intervention makes teleoperation problematic, and suggests a need for decentralized coordination schemes which feature collective decision-making by individual autonomous robots. Cost considerations when applied to large groups of searchers suggest a need for distributed coordination that uses shared data to overcome limited sensor precision.
Designing a robot team to search a sensate region for a specific target phenomenon involves numerous engineering tradeoffs among robot attributes and environmental variables. For example, battery-powered robots have a finite energy store and can search only a limited area before depleting it. Success in finding a target source with finite energy resources can also depend on other characteristics of the robot such as sensor accuracy and noise, and efficiency of the locomotive subsystem, and well as properties of the collective search such as the number of robots in the team, the use of shared information to reduce redundant search, and the team coordination strategy used to ensure a coherent search process.
Numerous team coordination strategies have been proposed. See, e.g., Cao et al. "Cooperative Mobile Robotics: Antecedents and Directions", Proceedings of IEEE/RSJ IROS (1995). Strategies for cooperative action encompass theories from such diverse disciplines as artificial intelligence, game theory and economics, theoretical biology, distributed computing and control, animal ethology, and artificial life. For example, Reynolds simulated the formation of flocks, herds, and schools in which multiple autonomous agents were driven away from obstacles and each other by inverse square law repulsive forces. See Reynolds "Flocks, Herds, and Schools", Computer Graphics, Volume 21 No. 4, pp. 25-34 (1987). Part of the motivation behind Reynolds' work is the impression of centralized control exhibited by actual bird flocks, animal herds, and fish schools, despite the fact that each agent (bird, animal, or fish) is responding only to its limited range local perception of the world.
Most current coordination strategies do not include a formal development of the system dynamics. See, e.g., Brooks "Intelligence Without Reason", Proceedings IJCAO-91 (1991); Misawa "Discrete-Time Sliding Mode Control: the Linear Case", Journal of Dynamic Systems, Measurement, and Control, Volume 119 (1997). Consequently, important system properties such as stability, reachability, observability, and robustness cannot be characterized. Many of the schemes rely on stable controls at a lower level and provide coordination at a higher level. The coordination is often heuristic and ad hoc.
Appropriate coordination strategies can be used in applications beyond teams of physical robots. For example, autonomous software agents, properly coordinated, can search information or trends in cyberspace or other electronic storage.
Accordingly, there is a need for a coordination method that can use shared information to reduce energy expended, compensate for noisy or deliberately misleading sensors, and allow robust collective searching.
The present invention comprises a decentralized coordination strategy called alpha-beta coordination that can use shared information to reduce energy expended, compensate for noisy or deliberately misleading sensors, and allow robust collective searching. The alpha-beta coordination strategy is a family of collective search methods that allow teams of communicating agents to implicitly coordinate their search activities through a division of labor based on self-selected roles and self-determined status. An agent can play one of two complementary roles. An agent in the alpha role is motivated to improve its status by exploring new regions of the search space. An agent in the beta role is also motivated to improve its status, but is conservative and tends to remain aggregated with other agents until alpha agents have clearly identified and communicated better regions of the search space. An agent can select its role dynamically based on its current status value relative to the status values of neighboring team members. Status can be determined by a function of the agent's sensor readings, and can generally be a measurement of source intensity at the agent's current location. An agent's decision cycle can comprise three sequential decision rules: (1) selection of a current role based on the evaluation of the current status data, (2) selection of a specific subset of the current data, and (3) determination of the next heading using the selected data. Variations of the decision rules produce different versions of alpha and beta behaviors that lead to different collective behavior properties.
Advantages and novel features will become apparent to those skilled in the art upon examination of the following description or may be learned by practice of the invention. The objects and advantages of the invention may be realized and attained by means of the instrumentalities and combinations particularly pointed out in the appended claims.
The accompanying drawings, which are incorporated into and form part of the specification, illustrate embodiments of the invention and, together with the description, serve to explain the principles of the invention.
FIG. 3(a,b) illustrates the learning curves for various alpha-beta ratios.
The present invention comprises a decentralized coordination strategy called alpha-beta coordination. The alpha-beta coordination strategy is a family of collective search methods that allow teams of communicating agents to implicitly coordinate their search activities through a division of labor based on self-selected roles and self-determined status. An agent can play one of two complementary roles. An agent in the alpha role is motivated to improve its status by exploring new regions of the search space. An agent in the beta role is also motivated to improve its status, but is conservative and tends to remain aggregated with other agents until alpha agents have clearly identified and communicated better regions of the search space. An agent can select its role dynamically based on its current status value relative to the status values of neighboring team members. Status can be determined by a function of the agent's sensor readings, and can generally be a measurement of source intensity at the agent's current location. An agent's decision cycle can comprise three sequential decision rules: (1) selection of a current role based on the evaluation of the current status data, (2) selection of a specific subset of the current data, and (3) determination of the next heading using the selected data. Variations of the decision rules produce different versions of alpha and beta behaviors that lead to different collective behavior properties.
Alpha-beta coordination strategy can be used to coordinate robots such as the robot 40 shown functionally in FIG. 4. Each robot 40 can comprise a position sensor 41 and a sensor 42 sensitive to the search goal. An alpha-beta coordination module 44 can access sensors 41, 42 communicate with other robots via communication module 45, and access additional information and memory 47 to control robot motion 43, 46. Partitioning a robot team into alpha and beta roles can produce a balance between exploration and exploitation. Alpha agents can expend energy exploring low status regions of the search space, but can communicate valuable state information to team members that can prevent costly re-exploration of low status regions. Alpha agents can seek to emulate and ultimately surpass the highest-performing team members and consequently can be more sensitive to the effects of transient noise and more susceptible to false status values. Beta agents can use energy conservatively by resisting transient influences and moving in a direct path to high status regions of the search space identified by alpha agents. Beta agents can resist noise and false status values by selective resampling and averaging of status data, but must rely on alpha agents to improve their status. Consequently, beta agents can be misled by noise and false status under some circumstances through second order effects if many alpha agents are misled.
The alpha-beta coordination strategy relies on the following assumptions:
1. agents have a reliable communications mechanism;
2. agents are positioned in the sensate region surrounding a target or goal;
3. the objective is to converge the agents on the target or goal;
4. a higher status value implies a higher probability that the target is located near the corresponding agent.
Alpha-beta agents search without centralized leadership or hierarchical coordination. The primary collective behavior of an alpha-beta team is to aggregate in a region of high intensity. Alpha-beta teams are robust to single point fail/stop failures in team members since agents can simply use the latest data transmitted by other team members without regard to the identity of the sender. Alpha-beta coordination can require minimal knowledge about the search environment. Agents need no prior assumptions about the nature of the intensity surface, its spatial coherence, gradient field, or any other analytical information. Individual agents need have no sensor memory, but without sensor memory can not locate a source alone. Accordingly, alpha-beta coordination is a reactive collective search strategy rather than a collaborative strategy. Agents are implicitly cooperative, and do not use explicit forms of collaboration.
A simple social metaphor can provide an intuitive but imprecise description of the basis for alpha-beta coordination. The cohesion of an alpha-beta society is based on a common normative goal: each agent is motivated to improve its social status by associating with other agents of higher status. Social status can be determined by a scalar function of the shared sensor data communicated by other agents. The only assumption underlying alpha-beta coordination is that the status function orders points in the search space according to the probability that a source is located at the point. On each decision cycle, each agent αi broadcasts its current social status as a scalar value si and a location vector νi to all other agents, and receives their status values in return. An agent attempts to improve its standing through emulation by moving to a region occupied by agents reporting superior status. This simple goal pressures agents to aggregate into groups in the region of highest known status.
Decision Rules
To determine its next destination, each agent can first compute a common ordered set V={νi}, i.e., the set of all location vectors, according to the linear ordering ≦ of agents provided by the status function. The agent can then partition V to divide its fellow agents into two relative castes. The alpha caste is the set A0 of all agents that have social standing superior to agent α0: A0={νk|sk>s0}. The beta caste B0 is the set of all agents that have social standing inferior to agent α0: B0={νk|sk≦s0}. The beta caste includes agents of equal status because an agent always seeks to improve its current status. The present invention includes various ways to use the alpha and beta castes to generate the agent's next heading. The vectors in A0can be used to influence the agent to move toward its members, creating a social pressure to improve by alpha-pull. The vectors in B0 can be used to influence the agent to move away from its members, creating a social pressure to improve by beta-push. A0, B0 and V can be used to create pressure to aggregate. Alpha-pull and beta-push are heuristic in nature and do not necessarily lead to average improvement in arbitrary environments. Different decision rules based in the data vectors in A0, B0, V and subsets thereof can lead to different global behaviors of alpha-beta teams.
Zero Information Decision Rules
A special case is when V=Bi. In this case every agent has identical status, corresponding to a zero information or maximum information entropy state. When a zero information state is detected, the team can disperse to broaden the search area by using beta-push (all agents are in each other's beta castes) to compute a heading that leads agents on the outer edges of the cohort region away from the centroid of the region. As the density of the team decreases, more agents are free to move away from the centroid. A minimum limit on team density can prevent the ultimate loss of coherence.
Lost Agent Decision Rules
Another special case is when V={ }. In this case the agent has lost contact with all other agents. The "lost" agent can remain stationary, conserving energy until another agent comes within range. Alternatively, the lost agent can begin a random search, a gradient search, or use the last known heading to determine a new heading in hopes of reestablishing contact with the team.
A general form of an alpha-beta update rule uses a linear combination of the vector data in V, as in equation 1. In equation 1, a(k) is a weighting vector derived from the application of a scalar function to the current status measurements corresponding to vectors in V. The nature of the function applied to the status measurement vector and the specific subset of vectors selected from V determine the group behavior exhibited by the team.
A0={ } Decision Rules
If A0={ } then agent α0 is the highest status agent in V (called a φ-alpha agent). Such an agent can not experience alpha-pull, and therefore has two possibilities for movement: (1) remain stationary, and (2) move according to beta-push. Remaining stationary allows the agent to serve as a stationary beacon for the rest of the team. This can be a conservative strategy since it conserves energy and ensures that the agent remains at the top of the heap, but does not immediately explore the region of highest intensity. The second option uses beta-push to move the agent away from lower status agents. This can be a more risky strategy because the agent's status might decrease, but it can provide more information to the team.
∀j:s0≦sj Decision Rules
If an agent determines that no other agent has a lower status, then the agent must use alpha-pull to motivate motion.
A0≠{ };∀j:s0>sj Decision Rules
Agents that determine that some other agents have higher status and some other agents have lower status can use both alpha-pull and beta-push to motivate motion. There are many possible decision rules for such agents. In general, selection of a decision rule involves determination of whether to be conservative or speculative in the attempt to improve status. A conservative decision rule can move the agent toward but not beyond higher status agents. A speculative decision rule can use other agents to determine a likely direction but can move beyond the other agents.
Equation 2 shows a speculative decision rule. In equation 2, ν*(k) is the location of a φ-alpha agent, and μ is a factor that provides pressure to the agent to move beyond the φ-alpha agent along a line passing through ν*(k) and νi(k).
Equation 3 shows a more conservative decision rule. In equation 3, VA(k) are all the members of Ai, and a(k) is the corresponding vector with elements given by equation 4. Movement according to the decision rule in equation 3 causes an agent to move toward the center of mass of the alpha set. Agents moving according to equation 3 will consequently tend to aggregate in the current region of highest status. The averaging of many alpha status positions can reduce the impact of noise and false high status readings.
It can be desirable to combine conservative and speculative decision rules. For example, the speculative decision rule in Equation 2 can be made more conservative by using a weighted average or composite of higher status agents instead of a single highest status agent (combining Equation 2's exploration with Equation 3's resistance to noise and false sensor readings).
As another example, an agent can select among decision rules based on its status relative to other agents. An agent can select a speculative decision rule like that in equation 2 if it is one of the m highest status agents. Such a self-selected alpha agent will then try to improve its status by moving toward and overshooting a current φ-alpha agent. Agents not in the m highest status agents can self-select a conservative decision rule like that in equation 3. Such agents will conserve energy by moving conservatively, but will still improve the overall status of the team. The description of collective dynamics and the sample results given below concern agents that self-select decision rules.
Collective Dynamics
The state-space formulation of the present invention comprises a system of linear, homogenous, time-varying difference equations of order N, where N is the instantaneous number of agents, as in equation 5.
Let C be the set of all agents, {α1,α2, . . . αN}, whose states are described by the data pairs {[zi(k),xi(k)]}, where zi(k) is the measurement vector and νi(k) is the position vector of the ith agent. Let z(k)=s(k)+w(k) where s(k) is the signal and w(k) is measurement noise and other uncertainties, and f(k) is a scalar function F(z(k)).
The rows of the system matrix at step k reflect the independent decisions of the N agents. At step k the system state in terms of the transition matrix and initial state is given by equation 6.
The behavior of Ψ(k,k0) as k→∞ determines the stability of the system. The rows of Ψ(k,k0) represent the transition function of the ith agent at step k. An agent's behavior will be stable if its corresponding transition function is bounded for all time. The φ-alpha and beta systems are asymptotically stable. The transition function of a φ-alpha agent at step k is given by equation 7.
Thus x100(k-1) is a stable node of the system at step k. The transition function of a beta agent is given by equation 8.
Equation 5 represents a stable system iff all ρj<1, the case that holds in an alpha-beta regime. The alpha agents that perform exploration can be stable, depending on the value of the convergence parameter μ. The transition function of an alpha agent is given by equation 9.
For exploration beyond the φ-alpha node at νφ, (k-1), μ>1.0 must hold. An asymptotically stable transition function occurs when 1.0<μ<2.0 and νφ(k-1) is the equilibrium point of the subsystem. The alpha subsystem is stable in the sense of Lyapunov when 1.0<μ<2.0 with an undamped oscillation about νφ(k-1) occurring for μ=2∅ Widespread exploration to escape a local extremum or plateaus can require that μ>2.0 resulting in a temporarily unstable alpha subsystem. The sufficient condition for Lyapunov stability of the time-varying alpha subsystem is given by equation 10.
In equation 10, |Ψ(k,k0) is the norm (largest eigenvalue) of the transition matrix and M is a finite constant. Eigenvalues with exponential growth are associated with alpha and φ-alpha points in the state space as described by equation 9. If μ>2.0 and equation 9 holds for a specific alpha agent for all k, then the alpha trajectory will be oscillatory and will increase without bound along the line L determined by L=ν100 (k-1)-να(k-1). Assuming the intensity Z(ν,t) is a Lyapunov function, i.e., Z'≦0 (the total derivative of Z is negative semidefinite) holds in the region of interest, and if νφ(k-1) is not the maximum intensity point along L, then the alpha agent will eventually become a φ-alpha agent upon discovering a point of higher intensity than νφ(k-1) along L. The agent will then be governed by equation 7 and will remain a stable node until surpassed by a higher intensity value.
Equation 9 will govern the agent's behavior as k approaches infinity if νφ(k-1) represents the point of maximum intensity (global maximum) in the region of interest. Accordingly, instability is possible only when the global maximum has been captured by at least one agent. Adaptive gain-scheduling techniques can reduce μ at convergence to stabilize the collective.
Sample Results
An alpha-beta coordination scheme was simulated in an ideal two dimensional world using ideal agents. The simulations illustrate the convergence and coordination properties of alpha-beta coordination. In the simulations, communication and sensing were noiseless, there were no obstacles, and agent movement at each step was bounded.
The target source was a radial emitter with exponential decay factor b and a uniform random noise component w, as in equation 11.
In equation 11 r is the radial distance from the origin. A metric of interest in the simulation is the mean-squared distance from the target, a measure of the team's learning rate and steady-state convergence error. For each simulation run, alpha-beta agents were initially positioned with the same distribution in the x-y plane. A control group comprising agents with identical starting points but with knowledge of the source location provided a baseline for the learning curve of the alpha-beta team. The simulation results confirm that the alpha-beta team can locate the target source.
The ratio R of alpha agents to beta agents can be important to an effective search. R intuitively corresponds to the ratio of exploration to exploitation, and can be fixed or can vary dynamically during a search. A critical mass of alpha agents can be needed to influence the beta agents to follow the alpha trajectory. A ratio of not less than 0.4 was needed for reliable search given μ=2∅ Maximum convergence rate and minimum steady-state mean-squared error occurred at R=0.5. FIG. 3(a,b) illustrates the learning curves for various alpha-beta ratios. Convergence was somewhat sensitive to μ, favoring greater values of μ at the expense of increased steady state mean-squared error. Very large values of μ can slow the convergence rate and lead to larger steady state errors.
The particular sizes and equipment discussed above are cited merely to illustrate particular embodiments of the invention. It is contemplated that the use of the invention may involve components having different sizes and characteristics. It is intended that the scope of the invention be defined by the claims appended hereto.
| Patent | Priority | Assignee | Title |
| 11179843, | Dec 22 2017 | Robert Bosch GmbH | Method for operating a robot in a multi-agent system, robot, and multi-agent system |
| 11198214, | May 15 2018 | Robert Bosch GmbH | Method for operating a robot in a multi-agent system, robot and multi-agent system |
| 6577906, | Aug 05 1999 | National Technology & Engineering Solutions of Sandia, LLC | Distributed optimization system and method |
| 6662169, | Apr 12 2001 | Controlling the self learning behavior of an autonomous agent | |
| 6826431, | Aug 05 1999 | National Technology & Engineering Solutions of Sandia, LLC | Distributed optimization system |
| 7171441, | Sep 07 2000 | Fujitsu Limited | Virtual communication channel and virtual private community, and agent collaboration system and agent collaboration method for controlling the same |
| 7908040, | Jul 15 2004 | Raytheon Company | System and method for automated search by distributed elements |
| 8112176, | Jan 30 2008 | System for self-organizing mobile robotic collectives | |
| 8527247, | Mar 20 2008 | National Technology & Engineering Solutions of Sandia, LLC | Nonlinear power flow feedback control for improved stability and performance of airfoil sections |
| 8930034, | Mar 21 2007 | National Technology & Engineering Solutions of Sandia, LLC | Computing an operating parameter of a unified power flow controller |
| Patent | Priority | Assignee | Title |
| 5388189, | Dec 06 1989 | NEXTIRAONE, LLC | Alarm filter in an expert system for communications network |
| 5778150, | Jul 01 1996 | International Business Machines Corporation; IBM Corporation | Flexible procedural attachment to situate reasoning systems |
| 5832467, | Sep 27 1995 | U.S. Philips Corporation | Behavior prediction for rule-based data processing apparatus |
| 5896297, | Apr 15 1996 | Robotube delivery system | |
| 5970482, | Feb 12 1996 | WILMINGTON TRUST, NATIONAL ASSOCIATION, AS COLLATERAL AGENT | System for data mining using neuroagents |
| 6021403, | Jul 19 1996 | Microsoft Technology Licensing, LLC | Intelligent user assistance facility |
| 6122572, | May 08 1995 | Rafael Armament Development Authority Ltd | Autonomous command and control unit for mobile platform |
| 6247002, | Dec 11 1996 | Sony Corporation | Method and apparatus for extracting features characterizing objects, and use thereof |
| Executed on | Assignor | Assignee | Conveyance | Frame | Reel | Doc |
| Jun 22 1999 | GOLDSMITH, STEVEN Y | Sandia Corporation | ASSIGNMENT OF ASSIGNORS INTEREST SEE DOCUMENT FOR DETAILS | 010349 | /0792 | |
| Jun 24 1999 | Sandia Corporation | (assignment on the face of the patent) | / | |||
| Feb 09 2004 | Sandia Corporation | ENERGY, U S DEPARTMENT | CONFIRMATORY LICENSE SEE DOCUMENT FOR DETAILS | 014394 | /0663 | |
| May 01 2017 | Sandia Corporation | National Technology & Engineering Solutions of Sandia, LLC | CHANGE OF NAME SEE DOCUMENT FOR DETAILS | 043889 | /0699 |
| Date | Maintenance Fee Events |
| Jan 18 2006 | REM: Maintenance Fee Reminder Mailed. |
| Feb 13 2006 | M1551: Payment of Maintenance Fee, 4th Year, Large Entity. |
| Feb 13 2006 | M1554: Surcharge for Late Payment, Large Entity. |
| Dec 10 2009 | M1552: Payment of Maintenance Fee, 8th Year, Large Entity. |
| Dec 04 2013 | M1553: Payment of Maintenance Fee, 12th Year, Large Entity. |
| Date | Maintenance Schedule |
| Jul 02 2005 | 4 years fee payment window open |
| Jan 02 2006 | 6 months grace period start (w surcharge) |
| Jul 02 2006 | patent expiry (for year 4) |
| Jul 02 2008 | 2 years to revive unintentionally abandoned end. (for year 4) |
| Jul 02 2009 | 8 years fee payment window open |
| Jan 02 2010 | 6 months grace period start (w surcharge) |
| Jul 02 2010 | patent expiry (for year 8) |
| Jul 02 2012 | 2 years to revive unintentionally abandoned end. (for year 8) |
| Jul 02 2013 | 12 years fee payment window open |
| Jan 02 2014 | 6 months grace period start (w surcharge) |
| Jul 02 2014 | patent expiry (for year 12) |
| Jul 02 2016 | 2 years to revive unintentionally abandoned end. (for year 12) |